Development And Validation Of The Diagnostic Predictive Model For The Risk Of Infection Presence In Acute-on-Chronic Liver Failure Patients

Objective:Developing and validating a diagnostic model based on laboratory indicators,followed by visualisation in nomogram providing an useful clinical decision-marking support tool for clinicians to assess the risk level of infection presence in patients with Acute-on-chronic liver failure(ACLF)for early diagnosis and control of infection to reduce ACLF mortality and improve the prognosis of ACLF patients.Methods:(1)From May 2013 to July 2021,a training set of 125 patients with ACLF from Shandong Provincial Hospital was included in this study,while 60 patients with ACLF from Jinan Center Hospital were used as an external validation set from October 2011 to April 2022.The infection status of each group determined the division into infected and uninfected.Data were collected from infected patients at the time of infection diagnosis and from uninfected patients at the time of admission,including clinical data on demographics,immunoinflammatory,liver function,metabolism,and coagulation-fibrinolysis.(2)Descriptive statistical analysis was performed on the training set,and covariance between variables was tested.By utilizing the Least Absolute Shrinkage and Selection Operator(Lasso)and logistic multivariate regression analysis,we screened independent diagnostic factors and construct multiple diagnostic models by combining them.(3)A comparison of the Receiver Operating Characteristic curve(ROC),Area Under the ROC(AUC),calibration curve,Hosmer-Lemeshow test,,clinical impact curve,Decision Curve Analysis(DCA),and Net Reclassification Index(NRI)were used to determine the optimal model visualized in a nomogram.The fivefold cross-validation technique was used for the internal validation of the model.The diagnostic efficiency of the model was validated in the external validation set.The optimal range and specificity of the model were evaluated by hierarchical analyses.Results:(1)Eleven variables were screened as candidate variables in the training set,which had significant differences between the two groups.Based on the severe multicollinearity among variables,Lasso regression and logistic regression were applied to screen white blood cell count(WBC),lymphocyte percentage(LYM%),blood urea nitrogen(BUN),and D-dimer as diagnostic factors.For improving predictive efficiency,we combined diagnostic factors to construct three simplified predictive models,including WLBD(WBC count,LYM%,BUN,Ddimer),WBD(WBC count,BUN,D-dimer),and WD(WBC count,D-dimer).(2)With AUROC 0.803(95%CI:0.723-0.883),the optimal model,WBD screened by various methods,was proposed as a novel diagnostic model for evaluating the risk level of infection presence in ACLF patients.Internal validation of the model showed the mean AUROC value of 0.78,and the diagnosis efficiency of WBD was obtained in an external validation set with the AUROC of 0.885(95%CI:0.786-0.984).(3)Stratified analysis was performed according to the etiology and stage of patients with ACLF.ACLF stages were stratified into two groups by early-stage(prophase and early stage,n=62)and late-stage(middle and end stage,n=63)subgroups with AUROC 0.873(76.9%,87%,95%CI:0.78-0.966),0.776(90.9%,56.7%,95%CI:0.657-0.895),respectively.In addition,we also divided the patients into 2 subgroups by ACLF etiology:HBV-related ACLF(n=94)with AUROC 0.791(96.4%,60.5%,95%CI:0.691-0.891),non-HBV-related ACLF(n=31)with AUROC 0.817(50%,100%,95%CI:0.668-0.966).What’s more,the AUROC of WBD in HBV-related ACLF patients in early-stage(prophase and early stage)subgroup is 0.905(100%,77.8%,95%CI:0.807-1.00)and the AUROC of WBD is 0.714(64%,75%,95%CI:0.552-0.876)in HBV-related ACLF patients in late-stage subgroup(middle and end stage).Conclusion:(1)Based on clinical and laboratory parameters,a new diagnostic model called WBD has been established,which provides a valuable clinical decision support tool for clinicians to assess the risk level of infection presence in ACLF patients.(2)The WBD model has a good diagnostic efficiency in assessing the risk of infection presence in ACLF patients with different etiologies and different stages of severity,and is best suited for early-stage ACLF patients with HBV-related ACLF.